A Bayesian Degree-Corrected Stochastic Block Model for Community Detection in Large Networks

• Activity Number: 61
• Type: Topic Contributed
• Date/Time: Sunday, August 3, 2014 : 4:00 PM to 5:50 PM
• Sponsor: IMS
• Author(s): Luis Carvalho*+
• Companies: Boston University
• Keywords: centroid estimation

Abstract

We discuss a degree-corrected version of a stochastic block model that aims to achieve a better resolution limit for community identification and is able to handle very large networks. We cast community detection as Bayesian generalized linear models and show that suitable priors are essential to adequately characterize community behavior. To estimate community assignments, we describe a new centroid estimator based on canonical projections and show that while this estimator is similar to Binder's estimator it can be obtained more efficiently. We further propose a latent specification based on "popularity" classes to handle large networks, and show that it identifies communities reliably and efficiently. We demonstrate the proposed model and inference on a number of classical network datasets and real-world large networks from the Stanford network analysis project. Finally, we offer a few concluding remarks on the model implementation and directions for future work. This is joint work with Lijun Peng and Matthew Morse.